Rspa 2017 0432

Kinematic dynamos in
rspa.royalsocietypublishing.org spheroidal geometries

D. J. Ivers
School of Mathematics and Statistics, University of Sydney, Sydney,
Research New South Wales 2006, Australia
Cite this article: Ivers DJ. 2017 Kinematic DJI, 0000-0001-6794-5824
dynamos in spheroidal geometries. Proc. R.
Soc. A 473: 20170432. The kinematic dynamo problem is solved numerically
for a spheroidal conducting fluid of possibly large
http://dx.doi.org/10.1098/rspa.2017.0432
Downloaded from https://royalsocietypublishing.org/ on 03 May 2023
aspect ratio with an insulating exterior. The solution

method uses solenoidal representations of the
Received: 20 June 2017 magnetic field and the velocity by spheroidal toroidal
Accepted: 1 September 2017 and poloidal fields in a non-orthogonal coordinate
system. Scaling of coordinates and fields to a spherical
geometry leads to a modified form of the kinematic
Subject Areas: dynamo problem with a geometric anisotropic
geophysics, astrophysics, computational diffusion and an anisotropic current-free condition
mathematics in the exterior, which is solved explicitly. The scaling
allows the use of well-developed spherical harmonic
techniques in angle. Dynamo solutions are found for
Keywords:
three axisymmetric flows in oblate spheroids with
magnetohydrodynamics, dynamo theory, semi-axis ratios 1 ≤ a/c ≤ 25. For larger aspect ratios
spheroid, axisymmetric flow strong magnetic fields may occur in any region of
the spheroid, depending on the flow, but the external
Author for correspondence: fields for all three flows are weak and concentrated
near the axis or periphery of the spheroid.
D. J. Ivers
e-mail: david.ivers@sydney.edu.au
1. Introduction
The generation of magnetic fields in the Earth, planets
and stars can be modelled to a good approximation
as self-exciting dynamos in electrically conducting
spherical or spherical shell cores with insulating
exteriors. Fortunately, spheres and spherical shells
are the bounded conductors with insulating exteriors,
for which kinematic dynamos [1–4], and dynamically
consistent dynamos [5–7], have the most mathematically
and numerically tractable, but still difficult, geometries
[8–13]. The simplest aspherical geometry is spheroidal
with one axis of arbitrary rotational symmetry. The
Earth’s core is a sphere at zeroth order and an oblate
spheroid at second order, with finer-scale core–mantle
boundary topography at higher orders. Spheroidal
dynamos allow pressure coupling between the core
and the mantle in dynamical dynamos and may
allow benchmarking of methods for more general
2017 The Author(s) Published by the Royal Society. All rights reserved.
topography. Spheroidal dynamos of large semi-axis ratio, i.e. highly aspherical, constitute an
2
important class of dynamo geometries for elliptic and disc galaxies, and possibly accretion discs;
E7-elliptical galaxies, for example, have major to minor semi-axis ratios a/c ∼ 3. Galactic models
rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

...................................................
are typically αω-dynamos and use spherical geometries or the thin disc approximation a/c → ∞.
For earlier works, see Stix [14], White [15] and Soward [16]. Only laminar dynamos are considered
herein but the method extends to mean-field types.
A spheroidal finite-element code has been developed by Wu & Roberts [17], finite-volume
codes by Ernst-Hullermann et al. [18] and Vantieghem et al. [19], and an axisymmetric spectral
element code by Guermond et al. [20]. The aim of this work is to separately develop pseudo-
spectral numerical solutions of the insulated kinematic dynamo problem (KDP), the rotating
thermal convection problem and the rotating magneto-convection problem in spheroids and
then to combine them to solve the dynamical dynamo problem. Pseudo-spectral codes generally
perform more efficiently in spherical benchmarks for a given accuracy than finite-element and
finite-volume codes [9–13], but there is no best approach. A hybrid method for the insulated
KDP is presented here, which is finite-difference in radius and Galerkin in angle. Galerkin or
Chebychev collocation methods can also be used in radius.
The dynamo action of a moving electrically conducting fluid, which occupies an oblate
spheroidal volume V in (Euclidean) space E3 , is considered. The velocity v of the fluid is
prescribed. The volume V has semi-axes a and c (a ≥ c), a rigid boundary Σ and is surrounded
by an insulating exterior E3 \V. The magnetic induction field B is governed by the equations
∂B
= ∇ 2 B + Rm ∇ × (v × B) in V; ∇ × B = 0 in E3 /V; ∇ · B = 0 in E3 ; (1.1)
∂t
where the problem has been non-dimensionalized using a typical length L, the magnetic diffusion
time L2 /η and a typical speed V of the flow. The magnetic diffusivity η is uniform and Rm := VL/η
is the magnetic Reynolds number. Useful choices for the length scale L are the major or minor
semi-axes of the spheroidal boundary Σ, the radius of the sphere of volume |V| = 43 π a2 c or the
radius of the circle of area π ac, which imply, respectively, a = 1 and 0 < c ≤ 1, c = 1 and 1 ≤ a < ∞,
c = 1/a2 and 1 ≤ a < ∞ or c = 1/a and 1 ≤ a < ∞. In cartesian coordinates (x, y, z) with the z-axis
aligned along the symmetry axis of the spheroid, the boundary Σ and outward (non-unit) normal
n are given by homoeoidal
x2 + y2 z2 (x1x + y1y ) z1z

+ = 1, n := + . (1.2)
a2 c2 a2 c2
The vector field n is defined and smooth everywhere in E3 , not only on Σ. Besides the magnetic
Reynolds number, a dimensionless shape parameter must be prescribed: the most useful are the
aspect ratio a/c or its reciprocal, the flattening f := 1 − c/a and the ellipticity
e := 1 − (c/a)2 .

The flattening and ellipticity are related by f = 1 − 1 − e2 and e = 2f − f 2 . Shape can also
−3/2
that a c = 1, then a/c = a = c
be parametrized bya semi-axis: if the volume is fixed so 2 3

ande = 1 − a−6 = 1 − c3 ; if a = 1 then a/c = 1/c and e = 1 − c2 ; or if c = 1, then a/c = a and
e = 1 − 1/a2 .
The tangential and normal components of B are continuous across Σ and B is self-exciting,
[n × B]Σ = 0, [n · B]Σ = 0, B = O(|r|−3 ) as |r| → ∞, (1.3)
where [ ]Σ denotes the jump outward across Σ and r = x1x + y1y + z1z is the radius vector from
the origin. The flow v satisfies the mass conservation equation and the impenetrable boundary
condition
∂ρ
+ Rm ∇ · (ρv) = 0, in V; n · v = 0, on Σ. (1.4)
∂t
If v is incompressible (1.4) reduces to
∇ · v = 0, in V. (1.5)
In §2 classes of oblate spheroidal toroidal and poloidal fields analogous to the spherical classes
3
are defined to reduce to two the number of fields necessary to represent a solenoidal magnetic
field or an incompressible velocity. A further non-essential simplification is to scale the KDP to a

...................................................
spherical geometry. The spherically scaled KDP produces equations with an anisotropic geometric
magnetic diffusion in V, analogous to the anisotropic turbulent diffusion models considered
by Phillips & Ivers [21,22], and an anisotropic current-free condition in the insulating exterior.
Numerical solution using spherical harmonic expansions in the homoeoidal angles (see (2.1))
analogous to the Bullard–Gellman equations [2] is described in §3. Results are given in §4 for
magnetic free-decay and compared with analytic solutions for aspect ratios 1 ≤ a/c ≤ 100 in the
axisymmetric case and 1 ≤ a/c ≤ 25 in the non-axisymmetric case; analytical free-decay rates are
derived for small ellipticity. Three critical spheroidal dynamos are found for 1 ≤ a/c ≤ 25 based on
three spherical axisymmetric dynamo flows of Dudley & James [23]. Alternative discretizations
and reduction to boundary conditions of the insulating exterior are considered in §5. Section 6 is
the conclusion.
2. A class of oblate spheroidal toroidal–poloidal fields

The geometry is parametrized using homoeoidal oblate (a ≥ c) spheroidal coordinates (r, θ, φ),
defined by
x = ar sin θ cos φ, y = ar sin θ sin φ and z = cr cos θ, (2.1)
where 0 ≤ r < ∞, 0 ≤ θ ≤ π and φ is the azimuthal angle. Note: r is not |r|. If (r, θ, φ) are the
coordinates of a point P(x, y, z), then P lies on the oblate spheroidal r-surface Σ(r), (x2 + y2 )/a2 +
z2 /c2 = r2 . Moreover, if Q is the projection of P parallel to the z-axis, onto the sphere of radius ar
escribed on the spheroid, then θ is the angle zOQ. The spheroid (1.2) corresponds to r = 1, i.e.
Σ(1) is Σ. The meridional sections of the r-surfaces are concentric homoeoidal ellipses of equal
ellipticity e. Their foci lie on the azimuthal circle of radius aer in the equatorial plane z = 0, so the
Σ(r) are not confocal, and aer → 0 as r → 0.
The coordinate system (r, θ , φ) is not orthogonal if a = c. There are two reciprocal sets
of basis vectors associated with (r, θ , φ), the covariant basis (e1 , e2 , e3 ) := (∂r r, ∂θ r, ∂φ r), and
the contravariant basis (e1 , e2 , e3 ) := (∇r, ∇θ , ∇φ). The bases (e1 , e2 , e3 ) and (e1 , e2 , e3 ) are
related by the usual reciprocity relations, ei = 12 Jijk ej × ek and ei = 12 J−1 ijk ej × ek , where J =
∂(x, y, z)/∂(r, θ , φ) = a2 c r2 sin θ is the Jacobian of the transformation (2.1), repeated indices
are summed and ijk , ijk are the unit rank-3 alternating tensors. The bases satisfy the bi-
j
orthogonality condition, ei · ej = δi . At any point on Σ(r), e1 is normal to Σ(r), and e2 and e3
are tangential; e1 is not parallel to e1 and hence not normal to Σ(r) except along principal
axes. In fact, n = ∇ 12 r2 = re1 . The vector element of surface area on Σ(r) is dΣ(r) = e2 ×
e3 dθ dφ = Je1 dθ dφ = n a2 c r sin θ dθ dφ and dΣ(r) = |n| a2 c r sin θ dθ dφ. The volume element
dV = a2 c r2 sin θ dr dθ dφ = |n|−1 dΣ(r)r dr. The operator ∇ = e1 ∂r + e2 ∂θ + e3 ∂φ .
The two bases have the important derivative properties, ∇ · (J−1 ei ) = 0 and ∇ × ei = 0. Thus,
the divergence of a vector field F in terms of its contravariant components, F = F1 e1 + F2 e2 + F3 e3 ,
is given by ∇ · F = J−1 {∂r (JF1 ) + ∂θ (JF2 ) + ∂φ (JF3 )}. Let er := e1 , eθ := e2 /r, eφ := e3 /r sin θ and B =
Br er + Bθ eθ + Bφ eφ , then B is solenoidal if and only if ∂r (r2 sin θ Br ) + ∂θ (r sin θ Bθ ) + ∂φ (r Bφ ) = 0.
By analogy with spherical polar coordinates, homoeoidal oblate spheroidal toroidal and poloidal fields
can be defined component-wise as the operators
◦2
∂φ T Λ S ∂θ ∂r (rS) ∂φ ∂r (rS)
T{T} := eθ − eφ ∂θ T and S{S} := −er + eθ + eφ , (2.2)
sin θ r r r sin θ
◦2
where sin2 θ Λ S = sin θ ∂θ (sin θ ∂θ S) + ∂φ ∂φ S, such that ∇ · T{T} = 0 and ∇ · S{S} = 0. The magnetic
field can thus be represented in E3 as a sum of oblate-spheroidal toroidal and poloidal fields,
B = T{T} + S{S}. For any T, S and r > 0 the fields T{T} and S{S} are orthogonal over Σ(r) in a
4
certain sense, see equation (2.9). The matching and self-exciting conditions (1.3) become
[S]Σ = 0, [∂r S]Σ = 0, [T]Σ = 0; S = O(r−2 ), T = O(r−3 ) as r → ∞.

(2.3)
...................................................
Note that, since a2 ce2 × e3 = er /r2 sin θ , a2 ce3 × e1 = eθ /r sin θ, a2 ce1 × e2 = eφ /r,

2 1 2 2 2 r∂φ S 2 3
T{T} = ∇ × a c T∇ r = ∇T × a cn and S{S} = ∇ × a c e − r sin θ∂θ Se .
2 sin θ
In particular, n · T{T} = 0. Typically, ∇ × T{T} = S{S} for any S and ∇ × S{S} = T{T} for any T.
Physically, an oblate spheroidal toroidal (poloidal) magnetic field does not generate, nor is
generated by, a purely oblate spheroidal poloidal (toroidal) electric current as in the spherical
case. If incompressible the velocity can also be represented as a sum of oblate spheroidal toroidal
and poloidal fields, v = T{t} + S{s}, with potentials t and s. (Only in ∂t and eγ t does t represent
the time.)
(a) Spherically scaled coordinates and fields

It is possible to proceed in the basis (er , eθ , eφ ), but it is simpler to scale the cartesian coordinates,
◦ ◦ ◦
x = x/a, y = y/a and z = z/c, and the components of the magnetic field and the velocity along
principal axes of the spheroid (1.2),
◦ ◦
∇ = L−1 · ∇,
◦ ◦ ◦
r = L · r = r · L, B = L · B = B · L, v=L·v=v·L and n = L · r, (2.4)
◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ 3
where r := x1x + y1y + z1z is the position vector in (x, y, z)-space E , the diagonal scale tensor
◦
L := a−1 (1x 1x + 1y 1y ) + c−1 1z 1z with inverse L−1 := a(1x 1x + 1y 1y ) + c1z 1z and the operator ∇ :=
1x ∂x◦ + 1y ∂y◦ + 1z ∂z◦ . The transformation is not simply a (passive) change of variable on the
◦ 3
manifold E3 , since the same cartesian unit vectors are used in E : (1x◦ , 1y◦ , 1z◦ ) = (1x , 1y , 1z ). The scale
tensor L can be separated into isotropic and anisotropic parts,

1 f
L= I+ 1z 1z and L−1 = a(I − f 1z 1z ), (2.5)
a 1−f
◦ ◦2 ◦ ◦ ◦ ◦ ◦
where I is the identity tensor. Note |r| = r and Λ = Λ · Λ where Λ := r × ∇. The stretching
◦ ◦ 3 ◦
transforms the r-surface Σ(r) from a spheroid into a sphere Σ(r) of radius in E ; the image V
of the conducting fluid region V under the transformation is the sphere r < 1 and the image
◦ ◦ ◦ ◦
of its boundary Σ is the unit sphere Σ = Σ(1) with normal n = r. The volume element dV =
◦ ◦ ◦ ◦ ◦ ◦ ◦
∂(x, y, z)/∂(x, y, z)dx dy dz = a c dV. Poincaré [24] and Bryan [25] used this device in the study
2
of the equilibrium of rotating, self-gravitating homogeneous liquids. In general, equations (2.4)

imply ⎫
◦ ◦ ◦ ◦ ◦
∇ · B = ∇ · B, ∇ · v = ∇ · v; B · ∇ = B · ∇ ⎬
◦
(2.6)
v · ∇ = v · ∇; n · B = r · B, n · v = r · v. ⎭
◦ ◦ ◦ ◦ ◦ ◦
and
◦3
The variables (r, θ , φ) introduced in (2.1) are spherical polar coordinates in E . The basis
(er , eθ , eφ ) in E3 is mapped to the orthonormal spherical polar basis (1r , 1θ , 1φ ) = L · (er , eθ , eφ )
◦3 ◦ ◦ ◦ ◦
in E . Thus the scalar components of B = Br 1r + Bθ 1θ + Bφ 1φ and B with respect to (er , eθ , eφ )
◦ ◦ ◦
are equal, Br = Br , Bθ = Bθ and Bφ = Bφ , but different with respect to (1x , 1y , 1z ). Hence the oblate
spheroidal toroidal–poloidal representation for the magnetic field becomes
◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦
B = L · (T{T} + S{S}) = T{T} + S{S} = ∇×Tr + ∇×∇×Sr. (2.7)
Thus oblate spheroidal toroidal (poloidal) fields are scaled to spherical toroidal (poloidal) fields
◦ 3
in E ,
◦2
◦ ∂φ T ◦ Λ S ∂θ ∂r (rS) ∂φ ∂r (rS)
T{T} = 1θ − 1φ ∂θ T and S{S} = −1r + 1θ + 1φ , (2.8)
sin θ r r r sin θ
◦ ◦ ◦
The conditions S dΩ = 0 and T dΩ = 0, where dΩ := sin θ dθ dφ is the element of solid angle 5
◦ ◦ ◦
on Σ(r), are imposed to ensure uniqueness of S and T. For all T, S and r > 0, T{T} and S{S} are
◦
orthogonal on Σ(r),

...................................................

dΣ(r) ◦ ◦ ◦
T{T} · D · S{S} = a2 c r T{T} · S{S} dΩ = 0, (2.9)
|n|
where D := L · L is positive-definite. The T{T} and S{S} fields are orthogonal on Σ(r) in the sense
of the first integral. Also
◦ ◦ ◦ ◦ ◦2 ◦ ◦ ◦ ◦ ◦ ◦ ◦ 2
r · T{T} = 0, r · S{S} = −Λ S, ∇×T{T} = S{T} and ∇×S{S} = T{−∇ S}. (2.10)
◦ ◦ ◦ ◦ 2 ◦
Thus ∇×B = T{−∇ S} + S{T}. It will be assumed throughout that any quantity in E3 has a ringed
◦ 3
version in E , except for r, θ , φ, the toroidal–poloidal potentials T, S, t, s, the basis vectors 1x , 1y ,
1z and the derived quantity Ynm (θ , φ).
(b) The spherically scaled kinematic dynamo problem

It is shown in §2b(i)–(iii) that, if the coordinates and fields are scaled as in (2.4), the induction
equations (1.1a,c) are transformed using (2.6) to
◦
∂B ◦ 2 e2 ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦3
◦ = ∇ + ∂z◦ z◦ B + Rm ∇ × (v × B) in V; ∇ · B = 0 in E , (2.11)
∂t 1−e 2
◦ ◦
where t := t/a2 and Rm := a2 Rm are a scaled time and magnetic Reynolds number. Moreover, the
current-free condition (1.1b) becomes
◦ ◦ ◦ ◦ 3 ◦
∇×(B − e2 Bz 1z ) = 0, in E \V. (2.12)
The matching conditions at the boundary Σ and the self-exciting conditions as r → ∞ (1.3) imply
◦ ◦
[B]Σ◦ = 0; B = O(r−3 ), as r → ∞ (2.13)
◦
and conversely. Clearly, v acts as a dynamo for the magnetic field B, i.e. B grows, if and only v
◦
acts as a dynamo for B; critical flows and magnetic fields correspond.
The mass conservation equation and velocity boundary condition (1.4) are transformed using
(2.6) to
◦
∂ρ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦
◦ + Rm ∇ · (ρ v) = 0, in V; n · v = 0 on Σ; (2.14)
∂t
◦ ◦ ◦ ◦
defining ρ by requiring ◦ ρ dV =
δV ρ dV for any corresponding volumes δ V and δV, i.e.
δV
◦ ◦
ρ = a2 cρ. Hence the flow v is kinematically feasible if and only if v is kinematically feasible. In
◦ ◦ ◦ ◦
particular, v is incompressible if and only if v is incompressible; and n · v = 0 on Σ if and only if
◦ ◦
n · v = 0 on Σ. The no-slip velocity condition is also preserved under (2.4), v = 0 on Σ if and only
◦
if v = 0 on Σ, but the stress-free condition on v has a more complicated form for v.
(i) The magnetic induction equation

By (2.6) (even without ∇ · B = 0),
◦ ◦ ◦
L · ∇ × (v × B) = (L · v)∇ · B + B · ∇(L · v) − (L · B)∇ · v − v · ∇(L · B) = ∇ × (v × B).
Thus in scaled coordinates the dot product of the magnetic induction equation (1.1)(a) with L
yields
◦ ◦ ◦ ◦ ◦ ◦ ◦
∂t B = ∇ · (D · ∇ B) + Rm ∇ × (v × B), (2.15)
where, using (2.5), the geometric diffusion tensor D := L · L and its inverse can be written as
e2
a2 D = I + 1z 1z and D−1 = a2 (I − e2 1z 1z ). (2.16)
1 − e2
Substituting (2.16) for D into (2.15) yields the spherically scaled magnetic induction equation
6
(2.11a). The new feature of equation (2.11), over the magnetic induction equation (1.1), is the
anisotropic geometric diffusion term which enhances the magnetic diffusion in the z-direction.

...................................................
(ii) The equations in the insulating exterior
The current-free condition (1.1b) in the insulating exterior E3 \V transforms to
◦ ◦ ◦ 3 ◦
∇ × B = ∇ · {(L × L−1 ) · B} = 0, in E \V. (2.17)
The equivalent but more useful equation (2.12) for the present formulation is derived here. Using
the form (2.5) of L,

◦ ◦ ◦ f
B = a(B − f Bz 1z ) and a∇ = ∇ + 1z ∂z◦ . (2.18)
1−f
Substituting these two expressions directly into (1.1b) gives
◦ ◦ ◦ ◦ f ◦ ◦3 ◦
∇ × B = ∇×B − f ∇×Bz 1z + 1z × ∂z◦ B = 0, in E \V. (2.19)
1−f
It is not obvious that the vector equation (2.19) is equivalent to two scalar equations in T and S.
However, observe that (2.19) implies
◦ ◦ ◦ 3 ◦
1z · ∇×B = 0 in E \V. (2.20)
◦ ◦ ◦
The ∇-divergence of the middle expression in (2.19) gives ∂z◦ (1z · ∇×B), which vanishes identically
if (2.20) is satisfied, so that (2.19) is in fact equivalent to just two scalar equations. This can be seen
more clearly by contracting the left equation in (2.19) with L, which yields
◦ ◦ ◦ ◦ f ◦ ◦ ◦
aL · ∇ × B = ∇×B − f ∇×Bz 1z + (1z × ∂z◦ B + 1z 1z · ∇×B). (2.21)
1−f
◦ ◦ ◦ ◦ ◦ ◦ ◦
Using the identity ∇×Bz 1z + 1z × ∂z◦ B + 1z 1z · ∇×B = ∇×B and 2f − f 2 = e2 reduces (2.21) to
◦ ◦ ◦
a(1 − f )L · ∇ × B = ∇×(B − e2 Bz 1z ). (2.22)
Thus, the current-free conditions (1.1b), (2.17) or (2.19) are equivalent to the current-free condition
(2.12). This equation differs in form from the original current-free condition (1.1b) by a subtractive
◦ ◦
O(e2 ) geometric anisotropy e2 ∇×Bz 1z .
By (1.1b) there exists a magnetic scalar potential Ψ with B = −∇Ψ in E3 \V and ∇ 2 Ψ = 0 by
◦ ◦ ◦
(1.1c). The potential Ψ and B are related by B = −D·∇Ψ where D is given by (2.16) and since
◦ ◦ ◦ ◦
∇ 2 = ∇·D·∇, Ψ satisfies the elliptic equation ∇·D·∇Ψ = 0. Hence
◦
◦ −(∇ + (e2 /(1 − e2 ))1z ∂z◦ )Ψ ◦ e22 ◦ 3 ◦

B= and ∇ + ∂ ∂ Ψ = 0,
◦ ◦ in E \V.
a2 1 − e2 z z
◦ ◦ ◦ ◦
In cylindrical polar coordinates, Bs = −∂s◦ Ψ/a2 , Bφ = −∂φ Ψ/sa2 , Bz = −∂z◦ Ψ/c2 . The equation
◦ ◦ ◦ ◦ ◦ ◦ ◦
∇ 2 Ψ = 0 implies ∇ 2 (∇Ψ )2 = 2∇∇Ψ :∇∇Ψ ≥ 0 and (∇·D·∇)(∇Ψ )2 = 2 tr[(∇ ∇Ψ )T ·D·∇ ∇Ψ ] ≥ 0.
Thus, the elliptic interior maximum principle [26] implies, noting (1.3)c, that maxE3 /V |B| occurs
◦ ◦
on Σ and max ◦ 3 ◦ |B| on Σ, although not necessarily at corresponding points.
E \V
(iii) The toroidal and poloidal equations 7
The solenoidal condition (2.11b) is satisfied by the toroidal–poloidal representation (2.7).
◦

Equations for the toroidal–poloidal potentials T and S in V follow from (2.11a) by using (2.10),
...................................................
◦ 2 ◦2 e2 ◦ ◦ ◦ ◦ ◦ ◦ ◦
(∂◦t − ∇ )Λ S = − r · ∂z◦ z◦ B − Rm r · ∇ × (v × B) (2.23)
1 − e2
and
◦ 2 ◦2 e2 ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦
(∂◦t − ∇ )Λ T = − r · ∇×∂z◦ z◦ B − Rm r · ∇×∇ × (v × B). (2.24)
1−e 2
◦3 ◦ ◦
In E \V, since the ∇-divergence of the left side of (2.12) is zero, the properties (2.10) of toroidal–
poloidal fields imply (2.12) is equivalent to the two scalar equations
◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦
r · ∇×(B − e2 Bz 1z ) = 0 and r · ∇×∇×(B − e2 Bz 1z ) = 0. (2.25)
◦ ◦ ◦
Expressions in T and S follow from the identities r · ∇ × f 1z = ∂φ f , (2.10), r · ∇×∇×(f 1z ) = r∂r ∂z◦ f −
◦ 2
r cos θ ∇ f , 1z = cos θ 1r − sin θ 1θ and ∂z◦ = cos θ ∂r − r−1 sin θ ∂θ . Thus,
◦ ◦ ◦ cos θ ◦ 2 sin θ
Bz = Br cos θ − Bθ sin θ = − Λ S− ∂θ ∂r (rS) − ∂φ T (2.26)
r r
and the left sides of (2.25) become

◦ ◦ ◦ ◦ ◦2 cos θ ◦ 2 sin θ
r · ∇×(B − e2 Bz 1z ) = −Λ T + e2 Λ ∂φ S + ∂θ ∂r (r∂φ S) + ∂φφ T (2.27)
r r
and
◦ ◦ ◦ ◦ ◦ ◦2 ◦ 2 ◦ 2
r · ∇×∇×(B − e2 Bz 1z ) = Λ ∇ S + e2 {r∂r (cos θ ∂r − sin θ ∂θ ) − r cos θ ∇ }

cos θ ◦ 2 sin θ
+ Λ S+ ∂θ ∂r (rS) + ∂φ T . (2.28)
r r
◦2
From properties of the operator Λ , (2.25) with (2.27) and (2.28) reduce to the spherical equations
◦ 2
T = 0 and ∇ S = 0 in the limit e → 0. In general, if e > 0 they couple non-axisymmetric T and
S fields in r > 1. Equations different from (2.25) are possible, e.g. for e > 0 equations (2.20) and
(2.26), noting (2.10), imply
cos θ ◦ 2 sin θ ◦ 2 ◦3 ◦
Λ T+ ∂θ ∂r (rT) − ∇ ∂φ S = 0, in E \V. (2.29)
r r
This is treated more fully in §5. By (2.27) and (2.28), conditions (2.3) and (2.25) are equivalent to
conditions (1.1b) and (1.3). However, it is not obvious that conditions (2.3) together with (2.12)
◦
are sufficient to guarantee that no current crosses Σ. Observe that the dot product of (2.12) with r
◦ ◦ ◦ ◦ ◦ ◦
yields a(1−f )n · ∇ × B = r · ∇×(B−e2 Bz 1z ), since n = L · r. Apart from factors a(1−f )|n| and |r| the
expressions on the left and first on the right are the normal components of the (dimensionless)
◦ ◦ ◦ ◦ ◦ ◦
current on Σ and Σ. The difficulty is the Bz term. From (2.3) and (2.27), [r · ∇×(B−e2 Bz 1z )]Σ◦ = 0.
Hence [n · ∇ × B]Σ = 0 and thus n · ∇ × B = 0 on Σ − .
◦
Finally, if v (equivalently v) satisfies the impenetrable boundary condition ((2.14)b) the velocity
poloidal potential s satisfies
s = 0 at r = 1. (2.30)
3. Numerical method
Equations (2.11), (2.12) and (2.13) are solved by discretizing the toroidal–poloidal equations (2.23),
(2.24), (2.25) with (2.27) and (2.28), using a Galerkin method in angle with spherical harmonics as
the basis functions and second-order finite differences in radius for simplicity, and solving subject
to (2.3). Other radial discretization methods are possible.
(a) The angular spectral equations 8
The magnetic toroidal–poloidal potentials are expanded in spherical harmonics

...................................................

m ◦
m m m (2n + 1)(n − m)!
f= fn (r, t)Yn (θ , φ) and Yn (θ , φ) := (−) Pn,m (cos θ)eimφ , (3.1)
n,m
(n + m)!
where f = T or S, and Pn,m is the Neumann-associated Legendre function
dm Pn (z) 1 dn 2
Pn,m (z) := (1 − z2 )m/2 and Pn (z) = (z − 1)n . (3.2)
dzm 2n n! dzn
◦2
Also (Ynm )∗ = (−)m Yn−m where an asterisk denotes complex conjugation; Λ Ynm = −λn Ynm where
◦ 2
λn := n(n+1); and ∇ f (r)Ynm = Dn f (r)Ynm for any function f (r), where Dn := r−2 [∂r (r2 ∂r ) − λn ]. The
spherical harmonics (3.1) form a complete orthonormal set with respect to the inner-product
◦ ◦
( f , g) := fg∗ dΩ/4π of complex scalar functions f and g on Σ(r). The n = 0 terms do not contribute
to the vector fields and are omitted. The boundary, matching and self-exciting conditions (2.3)
imply the spherical harmonic coefficients satisfy
−2
[Sm
n ]r=1 = 0, [∂r Sm
n ]r=1 = 0, [Tnm ]r=1 = 0; Sm
n = O(r ), Tnm = O(r−3 ) as r → ∞. (3.3)
For a self-exciting insulated spherical dynamo the behaviour of T as r → ∞ does not arise as
in (3.3) because T then vanishes in E3 \V.
(i) The conducting region

The spectral forms of the poloidal and toroidal induction equations (2.23) and (2.24) are
e2 ◦ ◦ ◦ ◦
(∂◦t − Dn )Sm
n = Snm {∂z◦ z◦ B} + Rm Snm {∇×(v × B)} (3.4)
1−e 2
and
e2 ◦ ◦ ◦ ◦
(∂◦t − Dn )Tnm = T m {∂ ◦ ◦ B} + Rm Tnm {∇×(v × B)}, (3.5)
1 − e2 n zz
where the (reduced) poloidal transform and the toroidal transform are defined by

1 ◦ 1 ◦
Snm {F} := Ynm∗ r · F dΩ and Tnm {F} := Ynm∗ Λ · F dΩ. (3.6)
4π λn 4π λn
The horizontal divergence equation is redundant. It is equivalent to applying the operator d1 to
the radial induction equation (2.23), where the operator dn := ∂r +n/r.
In time-stepping problems, the transforms (3.6) of the induction term are most efficiently
calculated numerically from values of F on a (θ, φ)-grid using fast-Fourier transforms in φ
and Gaussian quadrature in θ . However, in the eigen- and critical-value problems for steady
flows considered here the transforms are evaluated using angular spectral expansions [1,2]. The
transforms of the anisotropic diffusion terms are more problematic. The angular spectral forms of
the anisotropic magnetic diffusion terms in (3.4) and (3.5) have relatively few terms, and so are
preferable here to the numerical calculation of (3.6). The transforms are
◦ ◦ ◦ 2
Snm {∂z◦ z◦ B} = Lm
n (S, T) and Tnm {∂z◦ z◦ B} = Lm
n (T, −∇ S), (3.7)
where the differential operator Lm

n is defined by
n−2 m 2imcm
n
Lm
n (F, G) := cn−1,n D1−n,2−n Fm
n−2 + d1−n Gm
n−1
n n(n + 1)
2imcm
n+1 n+3 m
+ Cm m
n Dn Fn + dn+2 Gm
n+1 + c Dn+2,n+3 Fm (3.8)
n(n + 1) n + 1 n+1,n+2 n+2

and the operator Dn1 ,n2 := dn1 dn2 . The coefficients are defined by cm
n := (n2 −m2 )/(4n2 −1), cm
n1 ,n2 := 9
cm cm and
n1 n2
n−2 m n+3 m

Cm
...................................................
n := c + c . (3.9)
n n,n n + 1 n+1,n+1
Expansions (3.8) may be derived using the recurrence relations
cos θYnm = cm m m m
n+1 Yn+1 + cn Yn−1 and sin θ ∂θ Ynm = ncm m m m
n+1 Yn+1 − (n + 1)cn Yn−1 . (3.10)
The poloidal and toroidal equations (3.4) and (3.5) hold for n = 1 and n = 2, since the factor cmn in
the first two terms of (3.8) vanishes for n = 1 (forcing m = 1) and the first term clearly vanishes
for n = 2. The right side of equation (3.7)(b) contains third-order radial derivatives of Sm
n through
the terms d1−n Dn−1 Sm and d D S
n+2 n+1 n+1
m but only second-order derivatives of T m .
n
n−1
(ii) The insulating exterior

The angular spectral forms of the exterior equations (2.25) with (2.27) and (2.28) are now
◦ ◦ ◦ ◦2
derived. Each term is treated separately. The expansions of the first terms are r · ∇×B = −Λ T =
◦ ◦ ◦ ◦ ◦2 ◦ 2
n,m λn Tn Yn and r · ∇×(∇×B) = Λ ∇ S = − n,m λn Dn Sn Yn . The expansions of the second
m m m m
◦ ◦ ◦
term in each equation follow from r · ∇ × f 1z = ∂φ f = n,m imfnm Ynm , r · ∇×∇×(f 1z ) = r∂r ∂z◦ f −
◦ 2
r cos θ ∇ f and the formulae
⎫
m ⎪
r · ∇ × ∇ × (f 1z ) = {−(n + 1)cm m m m
n d1−n fn−1 + ncn+1 dn+2 fn+1 }Yn ⎪
⎪
⎬
n,m
(3.11)
m⎪
◦
and Bz = {−(n − 1)cm m m m m ⎪
n d1−n Sn−1 − imTn + (n + 2)cn+1 dn+2 Sn+1 }Yn ,⎪
⎭
n,m
which can be derived using relations (3.10). Thus, the spectral forms of (2.27) and (2.28) give
λn Tnm − e2 {−im(n − 1)cm m 2 m m m

n d1−n Sn−1 + m Tn + im(n + 2)cn+1 dn+2 Sn+1 } = 0 (3.12)
and

n−2 m imcmn m
Dn Sm
n +e
2
cn−1,n D1−n,2−n Sm
n−2 +
m
d1−n Tn−1 − Ĉn Dn Sm
n
n n

imcm
n+1 m n+3 m
− dn+2 Tn+1 + c Dn+2,n+3 Sm
n+2 = 0. (3.13)
n+1 n + 1 n+1,n+2
m
where Ĉn := (n + 1)cmn,n /n + ncn+1,n+1 /(n + 1). The horizontal divergence of (2.12) is equivalent
m
to applying d1 to the toroidal equation (2.25)(a).
(b) Radial discretization

The radial dependence is discretized using second-order finite differences. In the interior 0 ≤ r ≤
1− a uniform radial grid is used: rj = jh, j = 0 : J, with step size h = 1/J and rJ = 1−. In the exterior
1+ ≤ r < ∞ a uniform grid in the inverse radial coordinate ξ = 1/r is used: ξj = 1 + jhξ , j = 0 : Jξ ,
with step size hξ = −1/Jξ . Additional interior and exterior limiting grid points r = 1− and r = 1+
are allocated to the boundary r = 1 to simplify implementation of the finite-differencing. Thus
the full radial grid is rj = jh for j = 0 : J and rj = 1/[1 + (j − J)hξ ]/ for j = J+1 : J+Jξ with rJ = 1−,
rJ+1 = 1 and rJ+2 = 1+. The truncation levels J, Jξ and the related step sizes h, hξ can be specified
independently.
◦
The Sm m
n and Tn equations (3.4) and (3.5) are discretized at internal radial grid points in V; the Sn
m
equation is also discretized at the boundary grid point r = 1−. The Tn equation is not discretized
m
at r = 1−, since Tnm is continuous across the boundary by (3.3) but ∂r Tnm may be discontinuous
(figure 4). The Sm m
n and Tn equations (3.12) and (3.13) are discretized at the grid points in the 10
exterior E \V. There is one equation for each of the coefficient values Tnm (rj ) and Sm
3
n (rj ) at each
grid point rj .

...................................................
In the insulating exterior the change of variable from r to ξ in equations (3.12) and (3.13) is
ξ
accomplished with the identities, r∂r = −ξ ∂ξ , ∂r = −ξ 2 ∂ξ . Define the operators da := ∂ξ + aξ −1 ,
ξ ξ ξ ξ −2 ξ ξ
Da,b := da db and Dn := ∂ξ ξ − λn ξ , where a, b are constants. Then da = −ξ d−a , Da,b = ξ 4 D2−a,−b ,
2
ξ ξ ξ ξ
Dn = D−n,n and Dn = Dn+2,−n = ξ 4 D−n,n = ξ 4 Dn . Thus, the toroidal and poloidal equations
become
ξ ξ
λn Tnm + e2 {−im(n − 1)cm 2 m m 2 m 2 m
n ξ dn−1 Sn−1 + im(n + 2)cn+1 ξ d−n−2 Sn+1 − m Tn } = 0 (3.14)
and

ξ n−2 m ξ im m ξ m ξ
ξ 2 Dn Sm
n +e
2
c ξ 2 Dn+1,n−2 Sm
n−2 − c d Tm − Ĉn ξ 2 Dn Sm
n n−1,n n n n−1 n−1 n

im m ξ m n+3 m ξ
+ cn+1 d−n−2 Tn+1 + cn+1,n+2 ξ 2 D−n,−n−3 Sm
n+2 = 0. (3.15)
n+1 n+1
One-sided second-order right-boundary difference schemes are used at or near the boundary
in the interior. Left-boundary schemes are used in the exterior S-equation (3.15) at rJ+2 for the T
and S fields. Otherwise centred formulae are used.
(c) Eigen- and critical-value problems for steady flows

◦ ◦◦
For steady flows magnetic field solutions can be found with time dependence B ∝ eγ t = eγ t ,
◦
where t is the time and the growth rates are related by γ = a2 γ . Growth, criticality and decay are
◦ ◦ ◦
preserved by the scaling: Re γ = Re γ = 0 or Re γ Re γ > 0. Computations use γ and the magnetic
◦
Reynolds number Rm = a2 Rm , and the shape parameter a/c (or c/a) is prescribed. Values of the
growth rates and magnetic Reynolds numbers for different length scales can be derived from
◦
computations by scaling with the shape parameter. Since the growth rate γ depends only on the
◦ ◦
shape and Rm , γ (a, c, Rm ) = a−2 γ (c/a, Rm ). Thus, if the length scale L is the major or minor semi-
◦
axis, spherical radius or circle radius, so that a = 1, c = 1, a2 c = 1 or ac = 1, the associated growth

◦ ◦ ◦ ◦ ◦
rates γ and magnetic Reynolds numbers Rm are given by γ , γ (c/a)2 , γ (c/a)2/3 or γ (c/a), and Rm ,
◦ ◦ ◦ ◦
Rm (c/a)2 , Rm (c/a)2/3 or Rm (c/a), respectively. Setting a = 1 gives γ (c, Rm ) = γ (1, c, Rm ), and hence
γ (a, c, Rm ) = a−2 γ (1, c/a, a2 Rm ).
The spatial discretization approximates the problem by a matrix generalized eigenproblem
◦
Ax = γ Bx, where the matrix B is singular. The discretization produces banded matrices with
the narrowest band, in general, if all coefficients and their equations at each grid point are
blocked together. The matrix A is essentially block-pentadiagonal due to the centred difference
formulae with the one-sided formulae distorting this slightly. In principle, the generalized
◦ ◦
eigenvalues γ are the zeros of the characteristic polynomial det(γ B − A) = 0. If the matrices
A and B are N×N, the leading coefficient of this polynomial is det B, which is zero since B
is singular. Accordingly, there are less than N generalized eigenvalues. In fact, generally there
may be 0, 1, . . . , N−1 or infinitely-many eigenvalues if the characteristic polynomial vanishes
identically. Despite this theoretical aspect of the generalized eigenproblem no difficulty, such as
phantom eigenvalues, which may occur with the QZ-algorithm, has been observed in practice
with the methods used to determine selected generalized eigenvalues of the KDP: inverse
iteration, the implicitly restarted Arnoldi method and the power method with the conformal
transformation w = −(z−z1 )/(z−z2 ). The transformation maps the half-plane Re z > 0 to the
exterior of the circle |w| = 1, z1 → 0, z2 → ∞ and 12 (z1 +z2 ) → 1. The line Re z = c is mapped to the
circle (c−z2 )ww̄ + [c − 12 (z1 +z2 )](w+w̄) + c − z1 = 0, with centre [c − 12 (z1 +z2 )]/(c−z2 ) and radius
1
2 |z1 −z2 |/|c−z2 |. Applying this mapping to the spectrum of the generalized eigenproblem yields
a method for finding the eigenvalue of largest real part: apply the power method to the matrix
M := −I − (z2 −z1 )(A−z2 B)−1 B.
4. Results 11
Magnetic free-decay v = 0 provides an excellent test case, since the new features in equations

(2.11) and (2.12) are all active if the ellipticity e > 0. If e = 0, the conducting volume V is spherical
...................................................
with magnetic free-decay modes given by

−j2n−1,k t jn (jn−1,k r), −j2n,k t jn (jn,k r), r < 1,
Sm
n =e and Tnm =e (4.1)
jn (jn−1,k )r−n−1 0, r > 1,
where n ≥ 1, jn is the nth spherical Bessel function of the first kind and jn,k is the kth positive zero
of jn , i.e. jn (jn,k ) = 0 for n = 0, 1, . . . , k = 1, 2, . . .. There are n cells of non-zero S or T in latitude and
k cells in radius.
(a) Axisymmetric free-decay solutions

For e > 0, there exist analytical axisymmetric (m = 0) free-decay (R = 0) solutions in terms of
spheroidal wave functions using confocal spheroidal coordinates. These functions are difficult
to compute even in arbitrary precision. Further details of the solutions are given in Wu & Roberts
[17]. The spheroidal coordinates are also used in §5.
An axisymmetric magnetic field has the representation B = ∇ × (χ 1φ /s) + Bφ 1φ , where χ is
the magnetic flux function, and χ = −a2 cr sin θ ∂θ S and Bφ = −a ∂θ T. The fields χ and Bφ are odd
◦ ◦ ◦ ◦ ◦
(even) in the equator if S and T are even (odd). The associated field B = ∇×(χ1φ /s) + Bφ 1φ , where
χ = χ /a2 c. Assuming χ (r, t) = χ̂(r) eγ t and Bφ (r, t) = B̂φ (r)eγ t , the magnetic induction equation
◦
(1.1) with Rm = 0 decouples into two scalar equations of the same form, ∇ 2 u − 2s−1 ∂s u − γ u = 0,
with either u = χ̂ or u = B̂φ . This equation has separable solutions in confocal oblate spheroidal
coordinates (ξ , φ, η). The change to (ξ , φ, η) from cylindrical polar coordinates (s, φ, z) is given by

s = d (1 + ξ 2 )(1 − η2 ) and z = dξ η, d := a2 − c2 = ae. (4.2)
The boundary (1.2) is then given by the level surface ξ = c/d = c/ae, since x = s cos φ and y = s sin φ.
(1)
The separable solutions are of the form u = sΞ (ξ )H(η), where Ξ (ξ ) = R1n (−iσ , iξ ) and H(η) =
S1n (−iσ , η) are radial and angular oblate spheroidal wave functions of the first kind, degree n and
order 1 (note d differs by a factor 2, [27]) and σ 2 = −d2 γ . Since Bφ = 0 on the insulating boundary
ξ = c/ae, the azimuthal modes are
, η) eγnk t ,
(1) a
a a
Bφ = R1n (−iσnk , iξ )S1n (−iσnk (4.3)
(1)
where γnk a (a, c) := −(σ a )2 /(a2 − c2 ) and σ a is the kth positive zero of R (−iσ , ic/ae) = 0. The field
nk nk 1n
Bφ is odd (even) in the equator if n is even (odd) and vanishes in the exterior. The growth
a (1, c) = γ◦ a (c) for a = 1 (suppressing R = 0) were calculated using the spheroidal wave
rates γnk nk m
◦
function package in M ATHEMATICA 8 [28]. The functions γ ank (c/a) for c/a = 0.01(0.01)0.1(0.05)1
(figure 1) are shown in appendix A table 1 with n = 1, 2, 3, k = 1 and n = 1, k = 2. The relative
errors in the numerical eigenvalues are shown in figure 2.
The meridional solutions are more complicated due to the matching conditions on χ at the
conducting boundary ξ = c/ae. In the exterior ξ > c/ae, χ → 0 as ξ → ∞, so the separable solutions
(2)
are Ξ = R1n (0, iξ ) = Q1n (iξ ) and H = S1n (0, η)P1n (η), where Qm n is an associated Legendre function
of the second kind,
dm
Qm 2
n (x) = (x − 1)
m/2
Qn (x) and
dxm
n
1 x+1 1
Qn (x) = Pn (x) log − Pk−1 (x)Pn−k (x)
2 x−1 k
k=1
z 12
2

...................................................
10 s
Figure 1. Ellipses of semi-axis ratios a/c = 1, 1.25, 2, 3, 4, 5 with c = 2 and a/c = 10, 20, 25, 50, 100 with a = 10.
(1,1,20 400 200)

(1,1,40 400 200)
(1,1,40 200 200) (1,1,40 800 200)
(1,1,80 400 400) (1,2,20 400 200)

1 10–2
(1,1,160 400 400) (1,2,40 400 200)
(1,1,240 200 800) (1,2,40 800 200)
10–1 (2,1,40 200 200) (2,1,20 400 200)
(2,1,80 400 400) 10–3 (2,1,40 400 200)
(2,1,80 1600 400) (2,1,40 800 200)
10–2 (2,1,80 3200 400) (3,1,20 400 200)
(3,1,40 400 200)
10–4
10–3 (3,1,40 800 200)
10–4
10–5
10–5
10–6
10–6
10–7 10–7
0.01 0.05 0.10 0.5 1.0 0.01 0.05 0.10 0.5 1.0
c/a c/a
Figure 2. Relative errors of axisymmetric meridional γnkm (a) and azimuthal γnka (b) free-decay rates versus 0.01 ≤ c/a ≤ 1 at
different truncation levels (N, J, Jξ ) for the values of (n, k, N, J, Jξ ) shown. (Online version in colour.)
and Pk is the Legendre polynomial (3.2)b of degree k. To satisfy the matching conditions, linear
combinations of these modes are required,
⎧
⎪
⎪ (1)
R1n (−iσ , iξ ) c
⎪
⎪
⎪ C n S1n (−iσ , η) eγ t , ξ < ,
⎨ n (1)
R (−iσ , ic/ae) ae
1n
χ= (4.4)
⎪
⎪ Q1l (iξ ) 1
⎪
⎪ c
⎪
⎩ Ĉ l P (η) eγ t ,
1 (ic/ae) l
ξ> .
Ql
ae
l
Since χ and ∇χ are continuous across the conducting boundary,

(1)
Cn d1n
l (−iσ ) = Ĉl+1 , Cn ∂ξ (log R1n (−iσ , iξ )) = Ĉl+1 ∂ξ (log Q1l+1 (iξ )), at ξ = c/ae,
n n

using Smn (−iσ , η) = l≥0 dmn
l (−iσ )Pl+m (η), where the primed summation is over l = n − m mod 2
m
and Smn (0, η) = Pn (η). Note dj (0) = δln−m . This yields two homogeneous linear systems
m mn

n Aln (σ )Cn = 0, where l(≥ 0) and n−1(≥ 0) are either both odd or both even, and
(1) c
Aln (σ ) := {ξ ∂ξ (log R1n (−iσ , iξ )) − ξ ∂ξ (log Q1l (iξ ))}d1n
l (−iσ ), ξ= .
ae
For non-trivial solutions det[Aln (σ )] = 0, which determines σ . The χ -modes are even (odd)
in the equator if n is odd (even). In the spherical limit a/c → 1, a → 1, e → 0, ξ = c/ae → ∞,
√ √ √ √
σ ξ → −γ , ξ ∂ξ (log R1n (−iσ , iξ )) → −γ j n ( −γ )/jn ( −γ ) and ξ ∂ξ [log Q1l+1 (iξ )] → −l − 2, using
(1)
13
([27], eqn. (4.1.15) and [29], eqns (14.8.15), (14.3.10)),
⎫

(l + 2m)! mn
...................................................
−1
(1 + ξ −2 )1/2 dl (−iσ )jl+m (σ ξ )⎪
(1) ⎪
Rmn (−iσ , iξ ) = Nmn il+m−n ⎪
⎪
l! ⎬
l≥0
(4.5)
μ ⎪
⎪
−iμπ Qν (w) μ π 1/2 ⎪
⎪
and e = Qν (w) ∼ , as |w| → ∞, ⎭
Γ (ν + μ + 1) Γ (ν + 3/2)(2w)ν+1

where the normalization factor Nmn := l≥0 dmn l (−iσ )(l + 2m)!/l!. Hence as a/c → 1, Aln (σ ) →
√ √ √
−γ jn−1 −γ δln−1 /jn −γ and γ → −j2n−1,k , n = 1, 2, . . . , recovering the m = 0 poloidal
modes in (4.1). The zeros of det[Aln (σ )] = 0, for e > 0 and hence the axisymmetric meridional
growth rates γnk m (a, c) can thus be found by continuation from e = 0. By continuity the cell structure
of the spherical free-decay modes extends to e > 0. The growth rates γnk m (1, c) = γ◦ m (c), for a = 1
nk
(suppressing Rm = 0) were computed in M ATHEMATICA 8, but note the angular coefficients d1k l
◦ ◦m
are not intrinsic functions [28,30]. The decay rate functions γ m 1,1 (c/a) and γ 2,1 (c/a) are given in
appendix A table 1 for c/a = 0.01(0.01)0.1(0.05)1. The relative errors in the numerical eigenvalues
are shown in figure 2. The agreement with the analytic solutions is excellent, but truncation levels
must increase substantially (mostly r) for c/a = 0.01.
(b) Free-decay solutions for small ellipticity

For small ellipticity e 1, free-decay solutions can be expanded in powers of e2 . Only the growth
rate to first order is calculated here; the modes are derived elsewhere [31]. The degeneracy in the
n and Tn−1 splits if e > 0 and motivates the following expansion in e ,
decay rates of Sm m 2
m m
Tn−1 = Tn−1,0 + e2 Tn−1,2
m
+ ··· , Sm m 2 m
n = Sn,0 + e Sn,2 + · · · , γ = γ0 + e2 γ2 + · · · (4.6)
and
m m
Tn−1±2l = Tn−1±2l,2l e2l + · · · , Sm m 2l
n±2l = Sn±2l,2l e + · · · , l ≥ 1, (4.7)
where γ0 = −λ2 , Tn−1−2lm = 0 if l > (n−1−m)/2 and Sm n−2l = 0 if l > (n−m)/2. Let λ = jn−1,k and
suppress k below. The l = 0 terms are spherical free-decay modes: if r < 1, Sm n,0 = σ jn (λr) and
m
Tn−1,0 = iτ jn−1 (λr), where σ and τ are constants with typically σ τ = 0; and if r > 1, Sm n,0 =
σ jn (λ)r −n−1 and Tn−1,0 = 0. Substitute expansions (4.6) and (4.7) into the interior equations (3.4)
m
and (3.5), and the exterior equations (3.12) and (3.13). Several useful factors of 1 − e2 , which affect
only subdominant terms, are retained for simplicity.
Since Dn jn (λr) = γ0 jn (λr), equations (3.4) and (3.5) are satisfied to O(1). To O(e2 ) they give,
applying the recurrence relations for spherical Bessel functions, dn+1 jn (λr) = λjn−1 (λr) and
d−n jn (λr) = −λjn+1 (λr),

m i 2mcm
n m
(γ0 − Dn−1 )Tn−1,2 =− σ γ0 λ − (C n−1 γ0 − γ̃2 )τ jn−1 (λr) (4.8)
1 − e2 n(n − 1)
and
1 2mcmn
(γ0 − Dn )Sm
n,2 = (C m
n γ0 − γ̃2 )σ + τ λ jn (λr), (4.9)
1 − e2 n(n + 1)
where γ̃2 := γ2 (1 − e2 ) and Cm

n is given by (3.9). In the exterior r > 1, the O(e ) equations from (3.12)
2
and (3.13) yield
m im(n + 1)cmn m
Tn−1,2 = dn+1 Sm
n,0 = 0 and Dn Sm m
n,2 = Ĉn Dn Sn,0 = 0. (4.10)
(n − 1)n
Hence the boundary and matching conditions (3.3) imply the boundary conditions,
m
Tn−1,2 = 0, dn+1 Sm
n,2 = 0, at r = 1 − . (4.11)
–1.00 14

–1.02
...................................................
4
–1.04
g 2
g0 –1.06
3
1
–1.08
0
20
1
–1.10
0 0.1 0.2 0.3 0.4

e
Figure 3. Splitting of free-decay modes. Comparison of computed γ with (4.12) for small ellipticity 0 ≤ e ≤ 0.4. Solid (+) and
broken (−) curves are ± γ4,1m with averaged denominator. Symbols triangle, circle, plus, diamond, square are computed values
with (N, J, Jξ ) = (20, 800, 200) for m = 0, 1, 2, 3, 4 (the integer labels on the right). The + γ4,10 and + γ4,12 curves are different.
(Online version in colour.)
To find the growth rate multiply (4.8) by r2 jn−1 (λr), (4.9) by r2 jn (λr) and integrate r over [0, 1].
1
Two integrations by parts, the boundary conditions (4.11), (γ0 − Dp )jp (λr) = 0 and 0 r2 j2p (λr) dr = 0
yield the solvability conditions
2mcmn 2mcmn
σ γ0 λ − (Cm
n−1 γ0 − γ̃2 )τ = 0 and (Cm
n γ0 − γ̃2 )σ + τ λ = 0.
n(n − 1) n(n + 1)
If σ τ = 0, (Cm ± m
n−1 − γ̃2 /γ0 )(Cn − γ̃2 /γ0 ) = (2mcn ) /[(n − 1)n ]. Thus, γ = γn (e) + O(e ) where
m m 2 2 2 4

± m e2 γ̃2 γ̃2 1 m 1 (4mcm
n)
2
γn (e) := γ0 + , = (Cn−1 +Cm n)± n −Cn−1 ) +
(Cm m 2 > 0. (4.12)
1−e 2 γ0 2 2 (n −1)n2
2
If e > 0, + γnm (e) > − γnm (e). Since ± γn−m = ± γnm , we consider only m ≥ 0. If m = n there is one non-
trivial root, γ̃2 /γ0 = Cnn = (n + 3)/[(n + 1)(2n + 3)] and τ = 0. If m = 0, either σ = 0 and γ̃2 /γ0 =
C0n−1 , or τ = 0 and γ̃2 /γ0 = C0n . Thus, for n > 1 and k ≥ 1 (restoring k) there 2n + 1 modes: − γnk m
+ m − 0 +
for m = 0 : n − 1 and γnk for m = 0 : n; for n = 1 there are two modes, γ1,k and γ1,k . Figure 3
1
compares computed values of γ with ± γ4,1 m and the zero j

3,1 ≈ 6.98793 on 0 ≤ e ≤ 0.4, i.e. γ0 =
√
−48.8312 and 1 ≤ a/c ≤ 5/ 21 ≈ 1.09. Using an averaged sub-dominant denominator 1 − 12 e2
gives better agreement than either 1 − e2 or 1 for e near 0.4.
(c) Non-axisymmetric free-decay solutions

The n = 1, k = 1 modes are − γ1,1
◦
0 = −( 4 + 1 (a/c)2 )π 2 + O(e4 ) = γ m , shown in appendix A table 1
5 5 1,1
for 0.01 ≤ c/a ≤ 1 and +γ 1 = −( 35 + 25 (a/c)2 )π 2 + O(e4 ), shown in appendix A table 2 for 1 ≤
1,1
a/c ≤ 25. For a/c = 1.01, the leading term gives +γ 1 ≈ −9.94896 compared with the eigenvalue
1,1
−9.94890 computed with (N, J, Jξ ) = (40, 400, 200); for a/c = 1.1, + γ1,1
1 ≈ −10.699 compared with
the eigenvalue −10.695. Typical spherical harmonic coefficients are shown in figure 4. Since B = 0
◦ ◦
if and only if B = 0 at a point, a magnetically hidden dynamo occurs only if B = 0 in E3 /V. Unlike
the sphere the only hidden free-decay modes in a spheroid are the m = 0 toroidal modes due to
diffusive coupling.
(d) Dynamos with axisymmetric flows

The dynamo action of three axisymmetric flows is now considered for aspect ratios 1 ≤ a/c ≤ 25.
◦ ◦
These three flows are constructed by choosing as v in V three spherical flows of Dudley & James
0 0.3
15
0.2
–0.2

...................................................
0.1
–0.4 0
1 1
S1 T2
–0.6 –0.1
–0.2
–0.8
–0.3
–1.0 –0.4
0 0.5 1.0 1.5 2.0 0 0.5 1.0 1.5 2.0
0 0.04
–0.02 0.02
–0.04 0
1 1
S3 T4
–0.06 –0.02
–0.08 –0.04
–0.10 –0.06
0 0.5 1.0 1.5 2.0 0 0.5 1.0 1.5 2.0
r r
Figure 4. The S11 (r) and Tn1 (r) coefficient functions of + γ1,11 for n ≤ 4. The toroidal coefficients T21 , T41 are non-zero in the
insulating exterior r > 1 and their first derivatives have a jump discontinuity at r = 1. (Online version in colour.)
[23] which exhibit dynamo action, and transforming them to kinematically feasible spheroidal
◦
flows v in V (see the discussion following (2.14)). Axisymmetry of v is preserved under the
◦
transformation to v if the symmetry axes of v and the spheroid V are aligned. In cylindrical
◦ ◦
polar coordinates (s, φ, z) and (s, φ, z) with common unit vectors (1s , 1φ , 1z ), where s is the physical
◦ ◦ ◦ ◦ ◦
cylindrical radius, the scale transformation (2.4) is s = as, z = cz, vs = av s , vφ = av φ and vz = cv z . An
incompressible axisymmetric flow has the meridional-azimuthal representation v = vm + ωs1φ
◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦
where vm = ∇ × (ψ1φ /s), and v = vm + ω(s, z)s1φ where vm = ∇ × (ψ1φ /s). Hence the stream
◦ ◦
functions and differential rotations are related by ψ(s, z) = a2 cψ(s/a, z/c) and ω(s, z) = ω(s/a, z/c).
◦ ◦ ◦
The Dudley & James [23] spherical dynamo flows are of the form v = ∇ × (ψ1φ /r sin θ) +
◦
ωr sin θ 1φ , where is a tuning parameter [32]. The associated toroidal–poloidal velocity
◦ ◦ ◦ ◦
potentials t and s are given by ∂θ s = −ψ/r sin θ and ∂θ t = −ωr sin θ. If ψ 1 := r sin π r sin2 θ, ψ 2 :=
◦ ◦
3r2 sin π r sin θ cos θ , ω1 := r−1 sin π r, ω2 := 3 sin π r cos θ, the three flows are (ψ 1 , ω1 ), (ψ 2 , ω1 ),
2 ◦ ◦ ◦ ◦
◦ ◦
(ψ 2 , ω2 ) with = 0.17, 0.13, 0.14, respectively, i.e. in toroidal–poloidal form,
(a) s1 t1 -flow: s = t = sin π r cos θ , s01 = t01 = √1 sin π r, = 0.17;

3
(b) s2 t1 -flow: s = 32 r sin π rcos2 θ , t = sin π r cos θ, s02 = √1 r sin π r, t01 = √1 sin π r, = 0.13;
5 3
(c) s2 t2 -flow: s = t = 32 r sin π rcos2 θ , s02 = t02 = √1 r sin π r, = 0.14;
5
√ √
using Y10 = 3 cos θ and Y20 = 5(−1 + 3 cos2 θ )/2. The flows are shown in figure 5 for a/c = 5.
Since the velocities are axisymmetric the magnetic field decouples into toroidal–poloidal
0 0
m , Tm , Sm+1 , Tm+1 , . . . for m = 0, 1, 2, . . . (omit S0 , T0 ). The s2 t1 -flow is equatorially
chains Sm m m m
0 0 0 0
symmetric (ES), i.e. it has a chain of the form t1 , s2 , t3 , s4 , . . . , so the magnetic field decouples
y1 y2 sw1 sw2
1 16
z 0

...................................................
–1
0 2 4 0 2 4 0 2 4 0 2 4
s s s s
◦ ◦
Figure 5. Stream functions ψ1 , ψ2 and azimuthal velocities sω1 and sω2 for a/c = 5. (Online version in colour.)
(a) (b)
105
104
104
103
R°c –w° c
103
102
102
10
1 10 5 10 15 20 25
a/c a/c
◦ ◦
Figure 6. (a) Critical magnetic Reynolds numbers Rc and (b) angular frequency ωc versus a/c for the flows (top to bottom)
◦
s1 t1 , s2 t1 and s2 t2 (Rc only). (Online version in colour.)
into an equatorially antisymmetric (EA) field with the chain Tm m , Sm , T m , . . . or an ES field

m+1 m+2
with the chain Sm , T m , Sm , . . . . The s t - and s t -flows are not ES so the magnetic field has no
m m+1 m+2 1 1 2 2
EA/ES symmetry. Numerically, the coefficient chains are truncated at degree N. Critical dynamo
◦
solutions with Re γ = 0 = Re γ are computed. Results for the three flows are shown in figure 6 and
in appendix A table 2. For a sphere (a/c = 1), the critical magnetic Reynolds number and frequency
are consistent with Dudley & James [23]. The slopes of the magnetic Reynolds number plots for
◦ ◦
a/c ≥ 5 indicate Rc ∝ (a/c)2.0 for s1 t1 , and Rc ∝ (a/c)1.9 for s2 t1 and s2 t2 . Truncation levels (N, J, Jξ )
are: for the s1 t1 flow, (80, 800, 100) if a/c = 3, 4, (80, 800, 200) if 5 ≤ a/c < 10 and (80, 1600, 200) if
10 ≤ a/c ≤ 25; for the s2 t1 flow, (80, 800, 100) if a/c < 5 and (80, 800, 200) if a/c ≥ 5; and for the s2 t2
flow (80, 800, 200).
Magnetic field plots are constructed for the normal, tangential and azimuthal components
in the natural orthonormal basis (1n , 1t , 1φ ), B = Bn 1n + Bt 1t + Ba 1φ . Geometrically, 1n and
1t are normal and tangent to the r-surface at any point, and lie in the meridional plane.
This is a hybrid non-coordinate basis since 1n = e1 /|e1 | = n/|n|, 1t = e2 /|e2 | = eθ /|eθ | and
◦ ◦ ◦ ◦ ◦
1φ = e3 /|e3 | = e3 /|e3 | = eφ /|eφ |. Now |n|Bn = n · B = r · L · B = r · B = rBr , thus Bn = rBr /|n|, where
(a) Re B (b) Im B 17

...................................................
z 0
–1
1
z 0
–1
1
z 0
–1
0 2 4 6 8 10 0 2 4 6 8 10
s s
Figure 7. Critical magnetic fields for the s1 t1 flow with a/c = 10. Real (a) and imaginary (b) parts of Bn , Bt , Ba on φ = 0. Solid
(dashed) lines are positive (negative). (Online version in colour.)
(a) Bn Bt Ba
z 0
–1
(b)
z 0
–1
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
s s s
Figure 8. Critical magnetic fields for the s1 t1 flow with a/c = 5. Real (a) and imaginary (b) parts of of Bn , Bt , Ba . (Online version
in colour.)
◦ ◦ ◦ ◦ ◦
ac|n| = r a2 cos2 θ + c2 sin2 θ . Also, since B = Br er + Bθ eθ + Bφ eφ , Bt = 1t
· B = Br 1t · er + Bθ 1t ·
eθ , where 1t · er = e1 · eθ /|eθ | = (a2 − c2 ) sin θ cos θ/|eθ | and 1t · eθ = |eθ | = a2 cos2 θ + c2 sin2 θ.
◦ ◦ ◦ ◦ ◦3
Lastly, Ba = cBφ . The components Br , Bθ , Bφ are computed from T and S in E as in the spherical
case. The critical magnetic fields of the s1 t1 -flow are shown in figure 7 for a/c = 10 and figure 8 for
a/c = 5. Strong magnetic field is confined near the axis with very weak field towards the periphery
(a) Bn Bt Ba
18

1
...................................................
z 0
−1
(b)
1
z 0
−1
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
s s s
Figure 9. (a,b) Data as in figure 8 but for the s2 t1 flow. (Online version in colour.)
(a) (b) (c)
z 0
−1
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
s s s
Figure 10. Critical magnetic field for the s2 t2 flow with a/c = 5. (a),(b),(c): Bn , Bt , Ba , respectively. (Online version in colour.)
of the spheroid, both internally and externally. The fields are reminiscent of the spherical results
of Latter & Ivers [32] at much larger magnetic Reynolds numbers. The critical fields of the s2 t1 -
flow and s2 t2 -flow for a/c = 5 are shown in figures 9 and 10, respectively. Strong magnetic field
occurs toward the periphery for the s2 t1 -flow, both internally and externally. For the s2 t2 -flow,
internally the strongest magnetic field occurs near the axis but there is also strong field towards
the periphery; externally the strongest field occurs at the periphery.
5. Alternative methods for the insulating exterior

Two alternative solution methods for the insulating exterior are given in this section. The first
method eliminates Tnm from the system of exterior equations. Solving (3.12) for Tnm gives
ime2
Tnm = {−(n − 1)cm m m m
n d1−n Sn−1 + (n + 2)cn+1 dn+2 Sn+1 }, in r ≥ 1 − . (5.1)
λn − e2 m2
Observe from the matching conditions (3.3) on Tnm and Sm n that equation (5.1) holds at r = 1±. Also,
since |m| ≤ n and e2 < 1 < n(n + 1)/m2 , λn = e2 m2 , there are no zero divisors. There is a simpler
equation than (3.12) for m = 0, the spectral form of (2.20),
− (n − 1)cm m m m m
n d1−n Tn−1 + (n + 2)cn+1 dn+2 Tn+1 + imDn Sn = 0, (5.2)
which can be derived from (2.29) and used with (5.1). However, explicit elimination of Tn±1 m from
19
(5.2) or (3.13) using (5.1) gives

2 (n − 2)(n − 1) m (n − 1)(n + 1) m

...................................................
Dn S m
n + e cn−1,n D1−n,2−n Sm
n−2 − c Dn Sm
λn−1 − e m
2 2 λn−1 − e2 m2 n,n n

n(n + 2) m m (n + 2)(n + 3) m m
− c Dn S n + c D n+2,n+3 S n+2 = 0. (5.3)
λn+1 − e2 m2 n+1,n+1 λn+1 − e2 m2 n+1,n+2
The above equation, which is valid for all m, can be used in r > 1 together with (5.1) at r = 1−.
The second method replaces the system of equations for Tnm and Sm n in the exterior by a coupled
system of boundary conditions at r = 1−, derived from (5.1) and relations between the scalar,
toroidal and poloidal potentials Ψ , T and S. Laplace’s equation ∇ 2 Ψ = 0 separates in the confocal
oblate spheroidal coordinates (4.2). The exterior solution is
c
Ψ= Ψnm Qm m
n (iξ )Pn (η)e
imφ
, in ξ > . (5.4)
d
n≥1,m
The
homoeoidal
coordinates (2.1) and confocal coordinates (4.2) are related by s = ar sin θ =
d 1 + ξ 2 1 − η2 and z = c cos θ = dξ η. Solving for r and θ gives

2 d2 2 2 d2 2 2 c 1 + ξ 2 1 − η2
r = 2 (1 + ξ )(1 − η ) + 2 ξ η , tan θ = .
a c a ξη
The r- and ξ -surfaces coincide if and only if the first equation holds for all |η| ≤ 1. This is true
if and only if r2 /d2 = (1 + ξ 2 )/a2 = ξ 2 /c2 , i.e. ξ = c/ae and r = ξ d/c = 1. Thus only the boundary
spheroid Σ is both an r-surface and an ξ -surface.
The vector field n defined in (1.2) is normal to r-surfaces and can be written in the form
n = 1s r sin θ/a
+ 1z r cos θ/c. On
Σ, r = 1 and
n = 1s sin θ/a + 1z cos θ/c. The confocal ξ basis vector
eξ := ∂ξ r = d(ξ 1 − η2 1s + η 1 + ξ 2 1z )/ 1 + ξ 2 is normal to the ξ -surface at any point. In fact,
eξ = c dn on Σ where ξ = c/ae and η = cos θ. Thus, using (2.6), ∂ξ Ψ = eξ · ∇Ψ = −eξ · B = −c dn ·
◦ ◦ ◦2
B = −c dr · B = cdΛ S. Thus, from (5.4),

ic
Ψnm Qm
n = iaceλn Sm
n, (5.5)
ae
where the prime indicates the derivative. A second relation between Ψ , S and T is given by the
◦ ◦ ◦2 ◦
horizontal divergence of ∇Ψ = −D−1 · B. Using (2.16)(b) and Λ ≡ (r∇ h )2 ,
◦2 ◦ ◦ ◦ ◦2 ◦ ◦
Λ Ψ = −a2 r2 ∇ h · (B − e2 Bz 1z )h = −a2 Λ ∂r (rS) + e2 a2 r2 ∇ h · (Bz 1z )h . (5.6)
Substituting (5.4) and evaluating (5.6) on Σ (ξ = c/ae, η = cos θ and r = 1) gives

−2 m m ic
◦
∂r Sm
n + S m
n + a Ψ n Qn = e2 Hnm {Bz 1z }, (5.7)
ae
introducing the horizontal divergence transform defined by

1 ◦ ◦
Hnm {F} := − Ynm∗ r∇ h · Fh dΩ. (5.8)
4π λn
Eliminating Ψnm from (5.5) and (5.8) yields the boundary condition

m 2 (ic/ae)Qmn (ic/ae) ◦
∂r Sn + 1 + e λn Sm 2 m
n = e Hn {Bz 1z }. (5.9)
Qm
n (ic/ae)
As a/c → 1, e → 0 and the condition (5.9) reduces to the spherical poloidal insulating boundary
n + (n + 1)Sn /r = 0, since (Olver et al. [29], eqns (14.3.10), (14.8.15))
condition ∂r Sm m
μ
Qν (w) π 1/2
e−iμπ = Qμ
ν (w) ∼ , as |w| → ∞.
Γ (ν + μ + 1) Γ (ν + 3/2)(2w)ν+1
The horizontal divergence of f 1z for a scalar field f is given by r∇h · (f 1z )h = − sin θ ∂θ f − 2 cos θ f .
20
Thus by (3.10) the spherical harmonic expansion for the horizontal divergence of f 1z is given
◦
by r∇h · (f 1z )h = − n,m {(n + 1)cm
n fn−1 − ncn+1 fn+1 }Yn . Setting f = Bz and using (3.11) yields
m m m m

...................................................
◦ 1
Hnm {Bz 1z } = {(n + 1)cm m m m m m
n [−(n − 2)cn−1 d2−n Sn−2 − imTn−1 + (n + 1)cn dn+1 Sn ]
λn
− ncm m m m m m
n+1 [−ncn+1 d−n Sn − imTn+1 + (n + 3)cn+2 dn+3 Sn+2 ]}. (5.10)
◦ ◦ ◦
There is a third relation between Ψ , S and T given by the surface curl of ∇Ψ = −D−1 · B: if Λ :=
◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦
r × ∇, then 0 = Λ · ∇Ψ = −Λ · (D−1 · B) = −Λ · (B − e2 Bz 1z ), but this is the toroidal equation in
the exterior ((2.25)a).
6. Conclusion
The method described in §3 applies to mean field dynamos with a rank-2 tensor α-effect, which
are more relevant to galactic dynamo models. The magnetic induction equation (2.11a) becomes

◦ ◦ 2 e2 ◦ ◦ ◦ ◦ ◦ ◦ ◦
B + Rm ∇ × (v × B) + Rα L · ∇ · (L × α · L−1 · B).
◦
∂t B = ∇ + ∂ ◦◦
1 − e2 z z
The method also applies directly to the KDP in prolate spheroids, and extends easily to tri-axial
ellipsoids and to homoeoidal ellipsoidal shells (using vector and tensor spherical harmonics). The
method can also be applied to thermal convection and magnetoconvection problems. For non-
homoeoidal shells such as confocal shells, non-homoeoidal toroidal–poloidal fields are required,
but derivation of the toroidal–poloidal spectral equations is more difficult (for the confocal case,
see [33,34]). Work on time-stepping the KDP is in progress.
Data accessibility. The research does not contain any experimental data. The code implementing the original
computational methods is not accessible as the methods are fully described.
Competing interests. I declare I have no competing interests.
Funding. The research was fully supported by the School of Mathematics and Statistics at the University of
Sydney.
Appendix A. Free-decay rates and growth rates

◦ m ◦ a
Table 1. Axisymmetric meridional γ nk and azimuthal γ nk growth rates with weights c/a and (c/a)2 for 0.01 ≤ c/a ≤ 1.
◦ m ◦ m ◦ a ◦ a ◦ a ◦ a
c/a γ 1,1 c/a γ 1,2 c/a γ 1,1 (c/a)2 γ 1,2 (c/a)2 γ 2,1 (c/a)2 γ 3,1 (c/a)2
1.00 9.869604401 20.190728556 20.190728556 59.679515944 33.217461914 48.831193644
..........................................................................................................................................................................................................
0.95 9.573463520 20.055037754 18.613749809 55.143044929 31.359968460 46.156321167

..........................................................................................................................................................................................................
0.90 9.277209597 19.938272263 17.113443397 51.154471585 29.583789242 43.268233071

..........................................................................................................................................................................................................
0.85 8.980829306 19.843715162 15.689963175 47.785740476 27.888848717 40.094806581

..........................................................................................................................................................................................................
0.80 8.684306079 19.775447577 14.343430635 44.987859397 26.274897455 36.684608046

..........................................................................................................................................................................................................
0.75 8.387619214 19.738608214 13.073916808 42.616368122 24.741463026 33.181370561

..........................................................................................................................................................................................................
0.70 8.090742726 19.739761920 11.881417136 40.534934346 23.287792006 29.720337242

..........................................................................................................................................................................................................
0.65 7.793643821 19.787435038 10.765816420 38.654989408 21.912783614 26.388556676

..........................................................................................................................................................................................................
0.60 7.496280891 19.892913332 9.726839734 36.924907840 20.614917500 23.235623972

..........................................................................................................................................................................................................
0.55 7.198600817 20.071466941 8.763983471 35.314099582 19.392181612 20.289491743

..........................................................................................................................................................................................................
0.50 6.900535321 20.344296723 7.876418810 33.803139044 18.242011186 17.566210599

..........................................................................................................................................................................................................
(Continued.)
Table 1. (Continued.)
21
◦
m ◦
m ◦
a ◦a ◦a ◦a
c/a γ 1,1 c/a γ 1,2 c/a γ 1,1 (c/a)2 γ 1,2 (c/a)2 γ 2,1 (c/a)2 γ 3,1 (c/a)2

...................................................
0.45 6.601995965 20.741754519 7.062858522 32.378570791 17.161256187 15.074925162
..........................................................................................................................................................................................................
0.40 6.302867166 21.308935269 6.321380045 31.030259621 16.146199525 12.820193872

..........................................................................................................................................................................................................
0.35 6.002996288 22.115977381 5.649209918 29.750035017 15.192646483 10.802780638

..........................................................................................................................................................................................................
0.30 5.702179241 23.278482385 5.042510539 28.530990811 14.296089721 9.019441606

..........................................................................................................................................................................................................
0.25 5.400138973 25.002008642 4.496284019 27.367114899 13.451922975 7.462044036

..........................................................................................................................................................................................................
0.20 5.096492218 27.692128802 4.004589491 26.253081717 12.655647658 6.116720679

..........................................................................................................................................................................................................
0.15 4.790695731 29.213416496 3.561191068 25.184123285 11.903020017 4.964576007

..........................................................................................................................................................................................................
0.10 4.481954125 41.511793699 3.160335450 24.155938651 11.190124920 3.984883255

..........................................................................................................................................................................................................
0.09 4.419755597 44.576686595 3.084822548 23.954844700 11.051993860 3.807895437

..........................................................................................................................................................................................................
0.08 4.357380346 48.394890704 3.010777197 23.755194930 10.915281480 3.636879220

..........................................................................................................................................................................................................
0.07 4.294814582 53.282588634 2.938165403 23.556960350 10.779961450 3.471693435

..........................................................................................................................................................................................................
0.06 4.232043027 59.763884660 2.866954338 23.360112471 10.646007969 3.312200030

..........................................................................................................................................................................................................
0.05 4.169048686 68.776654491 2.797112297 23.164623292 10.513395759 3.158263906

..........................................................................................................................................................................................................
0.04 4.105812582 82.184711526 2.728608657 22.970465280 10.382100054 3.009752757

..........................................................................................................................................................................................................
0.03 4.042313418 104.306446505 2.661413830 22.777611361 10.252096592 2.866536917

..........................................................................................................................................................................................................
0.02 3.978527174 143.485205010 2.595499228 22.586034899 10.123361606 2.728489210

..........................................................................................................................................................................................................
0.01 3.914426594 266.247562876 2.530837220 22.651011290 9.995871812 2.595484809

..........................................................................................................................................................................................................
◦
Table 2. Critical magnetic Reynold’s number Rc versus a/c for the s1 t1 and s2 t2 flows with m = 1 magnetic fields, and the s2 t1
flow with an EA (no ES dynamo has been found) m = 1 magnetic field. For most aspect ratios (N, J, Jξ ) = (20, 200, 50). For the
◦
s2 t2 flow Imγ c = 0.
v 0 s1 t1 s2 t1 s2 t2
◦ ◦ ◦ ◦ ◦
+
a/c γ1,11 Rc Imγ c Rc Im γ c Rc
1 −π 2 155.585 −34.1440 94.9263 −18.2102 53.5624
..........................................................................................................................................................................................................
1.2 −11.592 194.942 −41.9226 102.3607 −18.9060 62.4924

..........................................................................................................................................................................................................
1.5 −14.689 270.313 −55.986 119.2005 −20.8543 79.1692

..........................................................................................................................................................................................................
2 −21.108 441.614 −85.738 158.6265 −25.6386 115.1571

..........................................................................................................................................................................................................
3 −38.223 957.444 −168.675 273.0472 −39.1726 216.5595

..........................................................................................................................................................................................................
4 −60.643 1701.689 −283.199 431.8008 −57.2683 356.0900

..........................................................................................................................................................................................................
5 −88.189 2671.595 −429.725 633.9955 −79.8498 533.0085

..........................................................................................................................................................................................................
6 −120.80 3865.436 −608.442 879.4256 −106.970 746.9163

..........................................................................................................................................................................................................
7 −158.45 5282.018 −819.416 1167.983 −138.665 997.5940

..........................................................................................................................................................................................................
8 −201.11 6920.498 −1062.669 1499.612 −174.961 1284.916

..........................................................................................................................................................................................................
9 −248.79 8780.251 −1338.204 1874.280 −215.874 1608.804

..........................................................................................................................................................................................................
10 −301.46 10860.64 −1646.07 2291.963 −261.412 1969.214

..........................................................................................................................................................................................................
20 −1102 43782.25 −6501.62 8832.98 −972.20 7578.5

..........................................................................................................................................................................................................
25 −1689 68486.7 −10141.2 13715.5 −1502.2 11750

..........................................................................................................................................................................................................
References 22
1. Elsasser WM. 1946 Induction effects in terrestrial magnetism. Part 1. Theory. Phys. Rev. 69,

106–116. (doi:10.1103/PhysRev.69.106)
...................................................
2. Bullard EC, Gellman H. 1954 Homogeneous dynamos and terrestrial magnetism. Phil. Trans.
R. Soc. Lond. A 247, 213–278. (doi:10.1098/rsta.1954.0018)
3. Gubbins D. 1973 Numerical solutions of the kinematic dynamo problem. Phil. Trans. R. Soc.
Lond. A 274, 493–521. (doi:10.1098/rsta.1973.0074)
4. Pekeris CL, Accad Y, Shkoller B. 1973 Kinematic dynamos and the earth’s magnetic field. Phil.
Trans. R. Soc. Lond. A 275, 425–461. (doi:10.1098/rsta.1973.0111)
5. Glatzmaier GA, Roberts PH. 1995 A three-dimensional convective dynamo with rotating
and finitely conducting inner core and mantle. Phys. Earth Planet. Int. 91, 63–75.
(doi:10.1016/0031-9201(95)03049-3)
6. Kageyama A, Sato T. 1995 Computer simulation of a magnetohydrodynamic dynamo. Phys.

Plasmas 2, 1421–1431. (doi:10.1063/1.871485)
7. Kuang W, Bloxham J. 1997 An Earth-like numerical dynamo model. Nature 389, 371–374.
(doi:10.1038/38712)
8. Jones CA. 2011 Planetary magnetic fields and fluid dynamos. Annu. Rev. Fluid Mech. 43,
583–614. (doi:10.1146/annurev-fluid-122109-160727)
9. Christensen UR et al. 2001 A numerical dynamo benchmark. Phys. Earth Planet. Int. 128, 25–34.
(doi:10.1016/S0031-9201(01)00275-8)
10. Jones CA, Boronski P, Brun AS, Glatzmaier GA, Gastine T, Miesch MS, Wicht J. 2011
Anelastic convection-driven dynamo benchmarks. Icarus 216, 120–135. (doi:10.1016/j.icarus.
2011.08.014)
11. Jackson A et al. 2014 A spherical shell numerical dynamo benchmark with pseudo-vacuum
magnetic boundary conditions. Geophys. J. Int. 196, 712–723. (doi:10.1093/gji/ggt425)
12. Marti P et al. 2014 Full sphere hydrodynamic and dynamo benchmarks. Geophys. J. Int. 197,
119–134. (doi:10.1093/gji/ggt518)
13. Matsui H et al. 2016 Performance benchmarks for a next generation numerical dynamo model.
Geochem. Geophys. Geosyst. 17, 1586–1607. (doi:10.1002/2015GC006159)
14. Stix M. 1975 The galactic dynamo. Astron. Astrophys. 42, 85–89.
15. White MP. 1978 Numerical models of the galactic dynamo. Astron. Nachr. 299, 209–216.
(doi:10.1002/asna.19782990407)
16. Soward AM. 1978 A thin disc model of the galactic dynamo. Astron. Nachr. 299, 25–33.
(doi:10.1002/asna.19782990104)
17. Wu C-C, Roberts PH. 2009 On a dynamo driven by topographic precession. Geophys.
Astrophys. Fluid Dyn. 103, 467–501. (doi:10.1080/03091920903311788)
18. Ernst-Hullermann J, Harder H, Hansen U. 2013 Finite volume simulations of dynamos in
ellipsoidal planets. Geophys. J. Int. 195, 1395–1405. (doi:10.1093/gji/ggt303)
19. Vantieghem S, Sheyko A, Jackson A. 2016 Applications of a finite-volume algorithm for
incompressible MHD problems. Geophys. J. Int. 204, 1376–1395. (doi:10.1093/gji/ggv527)
20. Guermond JL, Laguerre R, Léorat J, Nore C. 2009 Nonlinear magnetohydrodynamics in
axisymmetric heterogeneous domains using a fourier/finite element technique and an interior
penalty method. J. Comput. Phys. 228, 2739–2757. (doi:10.1016/j.jcp.2008.12.026)
21. Phillips CG, Ivers DJ. 2000 Spherical anisotropic diffusion models for the Earth’s core. Phys.
Earth Planet. Int. 117, 209–223. (doi:10.1016/S0031-9201(99)00098-9)
22. Phillips CG, Ivers DJ. 2003 Strong field anisotropic diffusion models for the Earth’s core. Phys.
Earth Planet. Int. 140, 13–28. (doi:10.1016/j.pepi.2003.07.004)
23. Dudley ML, James RW. 1989 Time-dependent kinematic dynamos with stationary flows. Proc.
R. Soc. Lond. A 425, 407–429. (doi:10.1098/rspa.1989.0112)
24. Poincaré H. 1885 Sur l’équilibre d’une masse fluide animée d’un mouvement de rotation. Acta
Math. 7, 259–380. (doi:10.1007/BF02402204)
25. Bryan GH. 1889 The waves on a rotating liquid spheroid of finite ellipticity. Phil. Trans. R. Soc.
Lond. A 180, 187–219. (doi:10.1098/rsta.1889.0006)
26. Protter MH, Weinberger HF. 1984 Maximum principles in differential equations. New York, NY:
Springer.
27. Flammer C. 2005 Spheroidal wave functions. New York, NY: Dover Publications.
28. Falloon PE, Abbott PC, Wang JB. 2003 Theory and computation of spheroidal wavefunctions.
J. Phys. A 36, 5477–5495. (doi:10.1088/0305-4470/36/20/309)
29. Olver FWJ, Lozier DW, Boisvert RF, Clark CW. 2010 NIST handbook of mathematical functions.
23
New York, NY: Cambridge University Press.
30. Emig T, Graham N, Jaffe RL, Kardar M. 2008 Casimir manipulations: The orientation

...................................................
dependence of fluctuation-induced forces. Phys. Rev. D 77, 025005. (doi:10.1103/PhysRevD.
77.025005)
31. Ivers DJ. 2017 Magnetic free-decay modes in spheroids of small ellipticity. Sydney, Australia: School
of Mathematics & Statistics, University of Sydney.
32. Latter H, Ivers D. 2010 Spherical single-roll dynamos at large magnetic reynolds numbers.
Phys. Fluids 22, 066601. (doi:10.1063/1.3453712)
33. Schmitt D, Jault D. 2004 Numerical study of a rotating fluid in a spheroidal container.
J. Comput. Phys. 197, 671–685. (doi:10.1016/j.jcp.2003.12.014)
34. Schmitt D. 2006 Numerical study of viscous modes in a rotating spheroid. J. Fluid Mech. 567,
399–414. (doi:10.1017/S0022112006002497)

Rspa 2017 0432

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Rspa 2017 0432

Uploaded by

Copyright:

Available Formats

Kinematic dynamos in

rspa.royalsocietypublishing.org spheroidal geometries

aspect ratio with an insulating exterior. The solution

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

x2 + y2 z2 (x1x + y1y ) z1z

[n × B]Σ = 0, [n · B]Σ = 0, B = O(|r|−3 ) as |r| → ∞, (1.3)

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

2. A class of oblate spheroidal toroidal–poloidal fields

x = ar sin θ cos φ, y = ar sin θ sin φ and z = cr cos θ, (2.1)

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

(a) Spherically scaled coordinates and fields

of the equilibrium of rotating, self-gravitating homogeneous liquids. In general, equations (2.4)

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

1z and the derived quantity Ynm (θ , φ).

(b) The spherically scaled kinematic dynamo problem

(i) The magnetic induction equation

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

the form (2.5) of L,

Substituting these two expressions directly into (1.1b) gives

◦ −(∇ + (e2 /(1 − e2 ))1z ∂z◦ )Ψ ◦ e22 ◦ 3 ◦

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

where f = T or S, and Pn,m is the Neumann-associated Legendre function

(i) The conducting region

where the differential operator Lm

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

(ii) The insulating exterior

λn Tnm − e2 {−im(n − 1)cm m 2 m m m

to applying d1 to the toroidal equation (2.25)(a).

(b) Radial discretization

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

(c) Eigen- and critical-value problems for steady flows

axis, spherical radius or circle radius, so that a = 1, c = 1, a2 c = 1 or ac = 1, the associated growth

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

(a) Axisymmetric free-decay solutions

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

(1,1,20 400 200)

(1,1,80 400 400) (1,2,20 400 200)

Since χ and ∇χ are continuous across the conducting boundary,

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

(b) Free-decay solutions for small ellipticity

where γ̃2 := γ2 (1 − e2 ) and Cm

and (3.13) yield

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

0 0.1 0.2 0.3 0.4

compares computed values of γ with ± γ4,1 m and the zero j

(c) Non-axisymmetric free-decay solutions

(d) Dynamos with axisymmetric flows

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

(a) s1 t1 -flow: s = t = sin π r cos θ , s01 = t01 = √1 sin π r, = 0.17;

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

into an equatorially antisymmetric (EA) field with the chain Tm m , Sm , T m , . . . or an ES field

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

(a) (b) (c)

5. Alternative methods for the insulating exterior

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

Substituting (5.4) and evaluating (5.6) on Σ (ξ = c/ae, η = cos θ and r = 1) gives

rspa.royalsocietypublishing.org Proc. R. Soc. A 473: 20170432

Appendix A. Free-decay rates and growth rates

0.95 9.573463520 20.055037754 18.613749809 55.143044929 31.359968460 46.156321167

0.90 9.277209597 19.938272263 17.113443397 51.154471585 29.583789242 43.268233071